add ifdefs for dev_bsize (almost working)

[unix-history] / usr / src / sys / ufs / ffs / ufs_inode.c

diff --git a/usr/src/sys/ufs/ffs/ufs_inode.c b/usr/src/sys/ufs/ffs/ufs_inode.c
index f5a5e4d..c72684c 100644 (file)
--- a/usr/src/sys/ufs/ffs/ufs_inode.c
+++ b/usr/src/sys/ufs/ffs/ufs_inode.c
@@ -1,21 +1,38 @@
-/*     ufs_inode.c     4.4     81/03/09        */
-
-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/mount.h"
-#include "../h/dir.h"
-#include "../h/user.h"
-#include "../h/inode.h"
-#include "../h/ino.h"
-#include "../h/filsys.h"
-#include "../h/conf.h"
-#include "../h/buf.h"
-#include "../h/inline.h"
-
-#define        INOHSZ  63
-#define        INOHASH(dev,ino)        (((dev)+(ino))%INOHSZ)
-short  inohash[INOHSZ];
-short  ifreel;
+/*
+ * Copyright (c) 1982, 1986 Regents of the University of California.
+ * All rights reserved.  The Berkeley software License Agreement
+ * specifies the terms and conditions for redistribution.
+ *
+ *     @(#)ufs_inode.c 7.1.1.1 (Berkeley) %G%
+ */
+
+#include "param.h"
+#include "systm.h"
+#include "mount.h"
+#include "dir.h"
+#include "user.h"
+#include "inode.h"
+#include "fs.h"
+#include "buf.h"
+#include "cmap.h"
+#ifdef QUOTA
+#include "quota.h"
+#endif
+#include "kernel.h"
+
+#define        INOHSZ  512
+#if    ((INOHSZ&(INOHSZ-1)) == 0)
+#define        INOHASH(dev,ino)        (((dev)+(ino))&(INOHSZ-1))
+#else
+#define        INOHASH(dev,ino)        (((unsigned)((dev)+(ino)))%INOHSZ)
+#endif
+
+union ihead {                          /* inode LRU cache, Chris Maltby */
+       union  ihead *ih_head[2];
+       struct inode *ih_chain[2];
+} ihead[INOHSZ];
+
+struct inode *ifreeh, **ifreet;

 
 /*
  * Initialize hash links for inodes
 
 /*
  * Initialize hash links for inodes


@@ -25,15 +42,29 @@ ihinit()
 {
        register int i;
        register struct inode *ip = inode;
 {
        register int i;
        register struct inode *ip = inode;

+       register union  ihead *ih = ihead;

 
 

-       ifreel = 0;
-       for (i = 0; i < ninode-1; i++, ip++)
-               ip->i_hlink = i+1;
-       ip->i_hlink = -1;
-       for (i = 0; i < INOHSZ; i++)
-               inohash[i] = -1;
+       for (i = INOHSZ; --i >= 0; ih++) {
+               ih->ih_head[0] = ih;
+               ih->ih_head[1] = ih;
+       }
+       ifreeh = ip;
+       ifreet = &ip->i_freef;
+       ip->i_freeb = &ifreeh;
+       ip->i_forw = ip;
+       ip->i_back = ip;
+       for (i = ninode; --i > 0; ) {
+               ++ip;
+               ip->i_forw = ip;
+               ip->i_back = ip;
+               *ifreet = ip;
+               ip->i_freeb = ifreet;
+               ifreet = &ip->i_freef;
+       }
+       ip->i_freef = NULL;

 }
 
 }
 

+#ifdef notdef

 /*
  * Find an inode if it is incore.
  * This is the equivalent, for inodes,
 /*
  * Find an inode if it is incore.
  * This is the equivalent, for inodes,


@@ -41,17 +72,19 @@ ihinit()
  */
 struct inode *
 ifind(dev, ino)
  */
 struct inode *
 ifind(dev, ino)

-dev_t dev;
-ino_t ino;
+       dev_t dev;
+       ino_t ino;

 {
        register struct inode *ip;
 {
        register struct inode *ip;

+       register union  ihead *ih;

 
 

-       for (ip = &inode[inohash[INOHASH(dev,ino)]]; ip != &inode[-1];
-           ip = &inode[ip->i_hlink])
+       ih = &ihead[INOHASH(dev, ino)];
+       for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw)

                if (ino==ip->i_number && dev==ip->i_dev)
                        return (ip);
        return ((struct inode *)0);
 }
                if (ino==ip->i_number && dev==ip->i_dev)
                        return (ip);
        return ((struct inode *)0);
 }

+#endif notdef

 
 /*
  * Look up an inode by device,inumber.
 
 /*
  * Look up an inode by device,inumber.


@@ -69,91 +102,162 @@ ino_t ino;
  *     "cannot happen"
  */
 struct inode *
  *     "cannot happen"
  */
 struct inode *

-iget(dev, ino)
-dev_t dev;
-ino_t ino;
+iget(dev, fs, ino)
+       dev_t dev;
+       register struct fs *fs;
+       ino_t ino;

 {
        register struct inode *ip;
 {
        register struct inode *ip;

+       register union  ihead *ih;

        register struct mount *mp;
        register struct buf *bp;
        register struct dinode *dp;
        register struct mount *mp;
        register struct buf *bp;
        register struct dinode *dp;

-       register int slot;
+       register struct inode *iq;

 
 loop:
 
 loop:

-       slot = INOHASH(dev, ino);
-       ip = &inode[inohash[slot]];
-       while (ip != &inode[-1]) {
-               if(ino == ip->i_number && dev == ip->i_dev) {
-                       if((ip->i_flag&ILOCK) != 0) {
+       ih = &ihead[INOHASH(dev, ino)];
+       for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw)
+               if (ino == ip->i_number && dev == ip->i_dev) {
+                       /*
+                        * Following is essentially an inline expanded
+                        * copy of igrab(), expanded inline for speed,
+                        * and so that the test for a mounted on inode
+                        * can be deferred until after we are sure that
+                        * the inode isn't busy.
+                        */
+                       if ((ip->i_flag&ILOCKED) != 0) {

                                ip->i_flag |= IWANT;
                                sleep((caddr_t)ip, PINOD);
                                goto loop;
                        }
                                ip->i_flag |= IWANT;
                                sleep((caddr_t)ip, PINOD);
                                goto loop;
                        }

-                       if((ip->i_flag&IMOUNT) != 0) {
-                               for(mp = &mount[0]; mp < &mount[NMOUNT]; mp++)
-                               if(mp->m_inodp == ip) {
-                                       dev = mp->m_dev;
-                                       ino = ROOTINO;
-                                       goto loop;
-                               }
+                       if ((ip->i_flag&IMOUNT) != 0) {
+                               for (mp = &mount[0]; mp < &mount[NMOUNT]; mp++)
+                                       if(mp->m_inodp == ip) {
+                                               dev = mp->m_dev;
+                                               fs = mp->m_bufp->b_un.b_fs;
+                                               ino = ROOTINO;
+                                               goto loop;
+                                       }

                                panic("no imt");
                        }
                                panic("no imt");
                        }

+                       if (ip->i_count == 0) {         /* ino on free list */
+                               if (iq = ip->i_freef)
+                                       iq->i_freeb = ip->i_freeb;
+                               else
+                                       ifreet = ip->i_freeb;
+                               *ip->i_freeb = iq;
+                               ip->i_freef = NULL;
+                               ip->i_freeb = NULL;
+                       }

                        ip->i_count++;
                        ip->i_count++;

-                       ip->i_flag |= ILOCK;
+                       ip->i_flag |= ILOCKED;

                        return(ip);
                }
                        return(ip);
                }

-               ip = &inode[ip->i_hlink];
-       }
-       if(ifreel < 0) {
+
+       if ((ip = ifreeh) == NULL) {

                tablefull("inode");
                u.u_error = ENFILE;
                return(NULL);
        }
                tablefull("inode");
                u.u_error = ENFILE;
                return(NULL);
        }

-       ip = &inode[ifreel];
-       ifreel = ip->i_hlink;
-       ip->i_hlink = inohash[slot];
-       inohash[slot] = ip - inode;
+       if (ip->i_count)
+               panic("free inode isn't");
+       if (iq = ip->i_freef)
+               iq->i_freeb = &ifreeh;
+       ifreeh = iq;
+       ip->i_freef = NULL;
+       ip->i_freeb = NULL;
+       /*
+        * Now to take inode off the hash chain it was on
+        * (initially, or after an iflush, it is on a "hash chain"
+        * consisting entirely of itself, and pointed to by no-one,
+        * but that doesn't matter), and put it on the chain for
+        * its new (ino, dev) pair
+        */
+       remque(ip);
+       insque(ip, ih);

        ip->i_dev = dev;
        ip->i_dev = dev;

+       ip->i_fs = fs;

        ip->i_number = ino;
        ip->i_number = ino;

-       ip->i_flag = ILOCK;
+       cacheinval(ip);
+       ip->i_flag = ILOCKED;

        ip->i_count++;
        ip->i_count++;

-       ip->i_un.i_lastr = 0;
-       bp = bread(dev, itod(ino));
+       ip->i_lastr = 0;
+#ifdef QUOTA
+       dqrele(ip->i_dquot);
+#endif
+#ifdef SECSIZE
+       bp = bread(dev, fsbtodb(fs, itod(fs, ino)), (int)fs->fs_bsize,
+           fs->fs_dbsize);
+#else SECSIZE
+       bp = bread(dev, fsbtodb(fs, itod(fs, ino)), (int)fs->fs_bsize);
+#endif SECSIZE

        /*
         * Check I/O errors
         */
        /*
         * Check I/O errors
         */

-       if((bp->b_flags&B_ERROR) != 0) {
+       if ((bp->b_flags&B_ERROR) != 0) {

                brelse(bp);
                brelse(bp);

+               /*
+                * the inode doesn't contain anything useful, so it would
+                * be misleading to leave it on its hash chain.
+                * 'iput' will take care of putting it back on the free list.
+                */
+               remque(ip);
+               ip->i_forw = ip;
+               ip->i_back = ip;
+               /*
+                * we also loose its inumber, just in case (as iput
+                * doesn't do that any more) - but as it isn't on its
+                * hash chain, I doubt if this is really necessary .. kre
+                * (probably the two methods are interchangable)
+                */
+               ip->i_number = 0;
+#ifdef QUOTA
+               ip->i_dquot = NODQUOT;
+#endif

                iput(ip);
                return(NULL);
        }
        dp = bp->b_un.b_dino;
                iput(ip);
                return(NULL);
        }
        dp = bp->b_un.b_dino;

-       dp += itoo(ino);
-       iexpand(ip, dp);
+       dp += itoo(fs, ino);
+       ip->i_ic = dp->di_ic;

        brelse(bp);
        brelse(bp);

-       return(ip);
+#ifdef QUOTA
+       if (ip->i_mode == 0)
+               ip->i_dquot = NODQUOT;
+       else
+               ip->i_dquot = inoquota(ip);
+#endif
+       return (ip);

 }
 
 }
 

-iexpand(ip, dp)
-register struct inode *ip;
-register struct dinode *dp;
+/*
+ * Convert a pointer to an inode into a reference to an inode.
+ *
+ * This is basically the internal piece of iget (after the
+ * inode pointer is located) but without the test for mounted
+ * filesystems.  It is caller's responsibility to check that
+ * the inode pointer is valid.
+ */
+igrab(ip)
+       register struct inode *ip;

 {
 {

-       register char *p1, *p2;
-       register int i;
+       while ((ip->i_flag&ILOCKED) != 0) {
+               ip->i_flag |= IWANT;
+               sleep((caddr_t)ip, PINOD);
+       }
+       if (ip->i_count == 0) {         /* ino on free list */
+               register struct inode *iq;

 
 

-       ip->i_mode = dp->di_mode;
-       ip->i_nlink = dp->di_nlink;
-       ip->i_uid = dp->di_uid;
-       ip->i_gid = dp->di_gid;
-       ip->i_size = dp->di_size;
-       p1 = (char *)ip->i_un.i_addr;
-       p2 = (char *)dp->di_addr;
-       for(i=0; i<NADDR; i++) {
-               *p1++ = *p2++;
-               *p1++ = *p2++;
-               *p1++ = *p2++;
-               *p1++ = 0;
+               if (iq = ip->i_freef)
+                       iq->i_freeb = ip->i_freeb;
+               else
+                       ifreet = ip->i_freeb;
+               *ip->i_freeb = iq;
+               ip->i_freef = NULL;
+               ip->i_freeb = NULL;

        }
        }

+       ip->i_count++;
+       ip->i_flag |= ILOCKED;

 }
 
 /*
 }
 
 /*


@@ -164,41 +268,59 @@ register struct dinode *dp;
  * truncate and deallocate the file.
  */
 iput(ip)
  * truncate and deallocate the file.
  */
 iput(ip)

-register struct inode *ip;
+       register struct inode *ip;
+{
+
+       if ((ip->i_flag & ILOCKED) == 0)
+               panic("iput");
+       IUNLOCK(ip);
+       irele(ip);
+}
+
+irele(ip)
+       register struct inode *ip;

 {
 {

-       register int i, x;
-       register struct inode *jp;
+       int mode;

 
 

-       if(ip->i_count == 1) {
-               ip->i_flag |= ILOCK;
-               if(ip->i_nlink <= 0) {
-                       itrunc(ip);
+       if (ip->i_count == 1) {
+               ip->i_flag |= ILOCKED;
+               if (ip->i_nlink <= 0 && ip->i_fs->fs_ronly == 0) {
+                       itrunc(ip, (u_long)0);
+                       mode = ip->i_mode;

                        ip->i_mode = 0;
                        ip->i_mode = 0;

+                       ip->i_rdev = 0;

                        ip->i_flag |= IUPD|ICHG;
                        ip->i_flag |= IUPD|ICHG;

-                       ifree(ip->i_dev, ip->i_number);
+                       ifree(ip, ip->i_number, mode);
+#ifdef QUOTA
+                       (void) chkiq(ip->i_dev, ip, ip->i_uid, 0);
+                       dqrele(ip->i_dquot);
+                       ip->i_dquot = NODQUOT;
+#endif

                }
                IUPDAT(ip, &time, &time, 0);
                }
                IUPDAT(ip, &time, &time, 0);

-               prele(ip);
-               i = INOHASH(ip->i_dev, ip->i_number);
-               x = ip - inode;
-               if (inohash[i] == x) {
-                       inohash[i] = ip->i_hlink;
+               IUNLOCK(ip);
+               ip->i_flag = 0;
+               /*
+                * Put the inode on the end of the free list.
+                * Possibly in some cases it would be better to
+                * put the inode at the head of the free list,
+                * (eg: where i_mode == 0 || i_number == 0)
+                * but I will think about that later .. kre
+                * (i_number is rarely 0 - only after an i/o error in iget,
+                * where i_mode == 0, the inode will probably be wanted
+                * again soon for an ialloc, so possibly we should keep it)
+                */
+               if (ifreeh) {
+                       *ifreet = ip;
+                       ip->i_freeb = ifreet;

                } else {
                } else {

-                       for (jp = &inode[inohash[i]]; jp != &inode[-1];
-                           jp = &inode[jp->i_hlink])
-                               if (jp->i_hlink == x) {
-                                       jp->i_hlink = ip->i_hlink;
-                                       goto done;
-                               }
-                       panic("iput");
+                       ifreeh = ip;
+                       ip->i_freeb = &ifreeh;

                }
                }

-done:
-               ip->i_hlink = ifreel;
-               ifreel = x;
-               ip->i_flag = 0;
-               ip->i_number = 0;
-       } else
-               prele(ip);
+               ip->i_freef = NULL;
+               ifreet = &ip->i_freef;
+       } else if (!(ip->i_flag & ILOCKED))
+               ITIMES(ip, &time, &time);

        ip->i_count--;
 }
 
        ip->i_count--;
 }
 


@@ -211,47 +333,38 @@ done:
  * i/o order so wait for write to complete.
  */
 iupdat(ip, ta, tm, waitfor)
  * i/o order so wait for write to complete.
  */
 iupdat(ip, ta, tm, waitfor)

-register struct inode *ip;
-time_t *ta, *tm;
-int waitfor;
+       register struct inode *ip;
+       struct timeval *ta, *tm;
+       int waitfor;

 {
        register struct buf *bp;
        struct dinode *dp;
 {
        register struct buf *bp;
        struct dinode *dp;

-       register char *p1, *p2;
-       register int i;
+       register struct fs *fs;

 
 

-       if((ip->i_flag&(IUPD|IACC|ICHG)) != 0) {
-               if(getfs(ip->i_dev)->s_ronly)
+       fs = ip->i_fs;
+       if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) != 0) {
+               if (fs->fs_ronly)

                        return;
                        return;

-               bp = bread(ip->i_dev, itod(ip->i_number));
+#ifdef SECSIZE
+               bp = bread(ip->i_dev, fsbtodb(fs, itod(fs, ip->i_number)),
+                       (int)fs->fs_bsize, fs->fs_dbsize);
+#else SECSIZE
+               bp = bread(ip->i_dev, fsbtodb(fs, itod(fs, ip->i_number)),
+                       (int)fs->fs_bsize);
+#endif SECSIZE

                if (bp->b_flags & B_ERROR) {
                        brelse(bp);
                        return;
                }
                if (bp->b_flags & B_ERROR) {
                        brelse(bp);
                        return;
                }

-               dp = bp->b_un.b_dino;
-               dp += itoo(ip->i_number);
-               dp->di_mode = ip->i_mode;
-               dp->di_nlink = ip->i_nlink;
-               dp->di_uid = ip->i_uid;
-               dp->di_gid = ip->i_gid;
-               dp->di_size = ip->i_size;
-               p1 = (char *)dp->di_addr;
-               p2 = (char *)ip->i_un.i_addr;
-               for(i=0; i<NADDR; i++) {
-                       *p1++ = *p2++;
-                       *p1++ = *p2++;
-                       *p1++ = *p2++;
-                       if(*p2++ != 0 && (ip->i_mode&IFMT)!=IFMPC
-                          && (ip->i_mode&IFMT)!=IFMPB)
-                               printf("iaddress > 2^24\n");
-               }
-               if(ip->i_flag&IACC)
-                       dp->di_atime = *ta;
-               if(ip->i_flag&IUPD)
-                       dp->di_mtime = *tm;
-               if(ip->i_flag&ICHG)
-                       dp->di_ctime = time;
-               ip->i_flag &= ~(IUPD|IACC|ICHG);
+               if (ip->i_flag&IACC)
+                       ip->i_atime = ta->tv_sec;
+               if (ip->i_flag&IUPD)
+                       ip->i_mtime = tm->tv_sec;
+               if (ip->i_flag&ICHG)
+                       ip->i_ctime = time.tv_sec;
+               ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
+               dp = bp->b_un.b_dino + itoo(fs, ip->i_number);
+               dp->di_ic = ip->i_ic;

                if (waitfor)
                        bwrite(bp);
                else
                if (waitfor)
                        bwrite(bp);
                else


@@ -259,154 +372,338 @@ int waitfor;
        }
 }
 
        }
 }
 

+#define        SINGLE  0       /* index of single indirect block */
+#define        DOUBLE  1       /* index of double indirect block */
+#define        TRIPLE  2       /* index of triple indirect block */

 /*
 /*

- * Free all the disk blocks associated
- * with the specified inode structure.
- * The blocks of the file are removed
- * in reverse order. This FILO
- * algorithm will tend to maintain
- * a contiguous free list much longer
- * than FIFO.
+ * Truncate the inode ip to at most
+ * length size.  Free affected disk
+ * blocks -- the blocks of the file
+ * are removed in reverse order.
+ *
+ * NB: triple indirect blocks are untested.

  */
  */

-itrunc(ip)
-register struct inode *ip;
+itrunc(oip, length)
+       register struct inode *oip;
+       u_long length;

 {
 {

-       register i;
+       register daddr_t lastblock;
+       daddr_t bn, lbn, lastiblock[NIADDR];
+       register struct fs *fs;
+       register struct inode *ip;
+       struct buf *bp;
+       int offset, osize, size, count, level;
+       long nblocks, blocksreleased = 0;
+       register int i;

        dev_t dev;
        dev_t dev;

-       daddr_t bn;
-       struct inode itmp;
+       struct inode tip;
+       extern long indirtrunc();

 
 

-       if (ip->i_vfdcnt)
-               panic("itrunc");
-       i = ip->i_mode & IFMT;
-       if (i!=IFREG && i!=IFDIR)
+       if (oip->i_size <= length) {
+               oip->i_flag |= ICHG|IUPD;
+               iupdat(oip, &time, &time, 1);

                return;
                return;

-
+       }
+       /*
+        * Calculate index into inode's block list of
+        * last direct and indirect blocks (if any)
+        * which we want to keep.  Lastblock is -1 when
+        * the file is truncated to 0.
+        */
+       fs = oip->i_fs;
+       lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
+       lastiblock[SINGLE] = lastblock - NDADDR;
+       lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
+       lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
+       nblocks = btodb(fs->fs_bsize);

        /*
        /*

-        * Clean inode on disk before freeing blocks
-        * to insure no duplicates if system crashes.
+        * Update the size of the file. If the file is not being
+        * truncated to a block boundry, the contents of the
+        * partial block following the end of the file must be
+        * zero'ed in case it ever become accessable again because
+        * of subsequent file growth.

         */
         */

-       itmp = *ip;
-       itmp.i_size = 0;
-       for (i = 0; i < NADDR; i++)
-               itmp.i_un.i_addr[i] = 0;
-       itmp.i_flag |= ICHG|IUPD;
-       iupdat(&itmp, &time, &time, 1);
-       ip->i_flag &= ~(IUPD|IACC|ICHG);
+       osize = oip->i_size;
+       offset = blkoff(fs, length);
+       if (offset == 0) {
+               oip->i_size = length;
+       } else {
+               lbn = lblkno(fs, length);
+               bn = fsbtodb(fs, bmap(oip, lbn, B_WRITE, offset));
+               if (u.u_error || (long)bn < 0)
+                       return;
+               oip->i_size = length;
+               size = blksize(fs, oip, lbn);
+               count = howmany(size, CLBYTES);
+               dev = oip->i_dev;
+               for (i = 0; i < count; i++)
+#ifdef SECSIZE
+                       munhash(dev, bn + i * CLBYTES / fs->fs_dbsize);
+#else SECSIZE
+                       munhash(dev, bn + i * CLBYTES / DEV_BSIZE);
+#endif SECSIZE
+               bp = bread(dev, bn, size);
+               if (bp->b_flags & B_ERROR) {
+                       u.u_error = EIO;
+                       oip->i_size = osize;
+                       brelse(bp);
+                       return;
+               }
+               bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
+               bdwrite(bp);
+       }
+       /*
+        * Update file and block pointers
+        * on disk before we start freeing blocks.
+        * If we crash before free'ing blocks below,
+        * the blocks will be returned to the free list.
+        * lastiblock values are also normalized to -1
+        * for calls to indirtrunc below.
+        */
+       tip = *oip;
+       tip.i_size = osize;
+       for (level = TRIPLE; level >= SINGLE; level--)
+               if (lastiblock[level] < 0) {
+                       oip->i_ib[level] = 0;
+                       lastiblock[level] = -1;
+               }
+       for (i = NDADDR - 1; i > lastblock; i--)
+               oip->i_db[i] = 0;
+       oip->i_flag |= ICHG|IUPD;
+       syncip(oip);

 
        /*
 
        /*

-        * Now return blocks to free list... if machine
-        * crashes, they will be harmless MISSING blocks.
+        * Indirect blocks first.

         */
         */

-       dev = ip->i_dev;
-       for(i=NADDR-1; i>=0; i--) {
-               bn = ip->i_un.i_addr[i];
-               if(bn == (daddr_t)0)
-                       continue;
-               ip->i_un.i_addr[i] = (daddr_t)0;
-               switch(i) {
+       ip = &tip;
+       for (level = TRIPLE; level >= SINGLE; level--) {
+               bn = ip->i_ib[level];
+               if (bn != 0) {
+                       blocksreleased +=
+                           indirtrunc(ip, bn, lastiblock[level], level);
+                       if (lastiblock[level] < 0) {
+                               ip->i_ib[level] = 0;
+                               free(ip, bn, (off_t)fs->fs_bsize);
+                               blocksreleased += nblocks;
+                       }
+               }
+               if (lastiblock[level] >= 0)
+                       goto done;
+       }

 
 

-               default:
-                       free(dev, bn);
-                       break;
+       /*
+        * All whole direct blocks or frags.
+        */
+       for (i = NDADDR - 1; i > lastblock; i--) {
+               register off_t bsize;

 
 

-               case NADDR-3:
-                       tloop(dev, bn, 0, 0);
-                       break;
+               bn = ip->i_db[i];
+               if (bn == 0)
+                       continue;
+               ip->i_db[i] = 0;
+               bsize = (off_t)blksize(fs, ip, i);
+               free(ip, bn, bsize);
+               blocksreleased += btodb(bsize);
+       }
+       if (lastblock < 0)
+               goto done;

 
 

-               case NADDR-2:
-                       tloop(dev, bn, 1, 0);
-                       break;
+       /*
+        * Finally, look for a change in size of the
+        * last direct block; release any frags.
+        */
+       bn = ip->i_db[lastblock];
+       if (bn != 0) {
+               off_t oldspace, newspace;

 
 

-               case NADDR-1:
-                       tloop(dev, bn, 1, 1);
+               /*
+                * Calculate amount of space we're giving
+                * back as old block size minus new block size.
+                */
+               oldspace = blksize(fs, ip, lastblock);
+               ip->i_size = length;
+               newspace = blksize(fs, ip, lastblock);
+               if (newspace == 0)
+                       panic("itrunc: newspace");
+               if (oldspace - newspace > 0) {
+                       /*
+                        * Block number of space to be free'd is
+                        * the old block # plus the number of frags
+                        * required for the storage we're keeping.
+                        */
+                       bn += numfrags(fs, newspace);
+                       free(ip, bn, oldspace - newspace);
+                       blocksreleased += btodb(oldspace - newspace);

                }
        }
                }
        }

-       ip->i_size = 0;
-       /*
-        * Inode was written and flags updated above.
-        * No need to modify flags here.
-        */
+done:
+/* BEGIN PARANOIA */
+       for (level = SINGLE; level <= TRIPLE; level++)
+               if (ip->i_ib[level] != oip->i_ib[level])
+                       panic("itrunc1");
+       for (i = 0; i < NDADDR; i++)
+               if (ip->i_db[i] != oip->i_db[i])
+                       panic("itrunc2");
+/* END PARANOIA */
+       oip->i_blocks -= blocksreleased;
+       if (oip->i_blocks < 0)                  /* sanity */
+               oip->i_blocks = 0;
+       oip->i_flag |= ICHG;
+#ifdef QUOTA
+       (void) chkdq(oip, -blocksreleased, 0);
+#endif

 }
 
 }
 

-tloop(dev, bn, f1, f2)
-dev_t dev;
-daddr_t bn;
+/*
+ * Release blocks associated with the inode ip and
+ * stored in the indirect block bn.  Blocks are free'd
+ * in LIFO order up to (but not including) lastbn.  If
+ * level is greater than SINGLE, the block is an indirect
+ * block and recursive calls to indirtrunc must be used to
+ * cleanse other indirect blocks.
+ *
+ * NB: triple indirect blocks are untested.
+ */
+long
+indirtrunc(ip, bn, lastbn, level)
+       register struct inode *ip;
+       daddr_t bn, lastbn;
+       int level;

 {
 {

-       register i;
-       register struct buf *bp;
+       register int i;
+       struct buf *bp, *copy;

        register daddr_t *bap;
        register daddr_t *bap;

-       daddr_t nb;
-
-       bp = NULL;
-       for(i=NINDIR-1; i>=0; i--) {
-               if(bp == NULL) {
-                       bp = bread(dev, bn);
-                       if (bp->b_flags & B_ERROR) {
-                               brelse(bp);
-                               return;
-                       }
-                       bap = bp->b_un.b_daddr;
-               }
+       register struct fs *fs = ip->i_fs;
+       daddr_t nb, last;
+       long factor;
+       int blocksreleased = 0, nblocks;
+
+       /*
+        * Calculate index in current block of last
+        * block to be kept.  -1 indicates the entire
+        * block so we need not calculate the index.
+        */
+       factor = 1;
+       for (i = SINGLE; i < level; i++)
+               factor *= NINDIR(fs);
+       last = lastbn;
+       if (lastbn > 0)
+               last /= factor;
+       nblocks = btodb(fs->fs_bsize);
+       /*
+        * Get buffer of block pointers, zero those 
+        * entries corresponding to blocks to be free'd,
+        * and update on disk copy first.
+        */
+       copy = geteblk((int)fs->fs_bsize);
+#ifdef SECSIZE
+       bp = bread(ip->i_dev, fsbtodb(fs, bn), (int)fs->fs_bsize,
+           fs->fs_dbsize);
+#else SECSIZE
+       bp = bread(ip->i_dev, fsbtodb(fs, bn), (int)fs->fs_bsize);
+#endif SECSIZE
+       if (bp->b_flags&B_ERROR) {
+               brelse(copy);
+               brelse(bp);
+               return (0);
+       }
+       bap = bp->b_un.b_daddr;
+       bcopy((caddr_t)bap, (caddr_t)copy->b_un.b_daddr, (u_int)fs->fs_bsize);
+       bzero((caddr_t)&bap[last + 1],
+         (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
+       bwrite(bp);
+       bp = copy, bap = bp->b_un.b_daddr;
+
+       /*
+        * Recursively free totally unused blocks.
+        */
+       for (i = NINDIR(fs) - 1; i > last; i--) {

                nb = bap[i];
                nb = bap[i];

-               if(nb == (daddr_t)0)
+               if (nb == 0)

                        continue;
                        continue;

-               if(f1) {
-                       brelse(bp);
-                       bp = NULL;
-                       tloop(dev, nb, f2, 0);
-               } else
-                       free(dev, nb);
+               if (level > SINGLE)
+                       blocksreleased +=
+                           indirtrunc(ip, nb, (daddr_t)-1, level - 1);
+               free(ip, nb, (off_t)fs->fs_bsize);
+               blocksreleased += nblocks;

        }
        }

-       if(bp != NULL)
-               brelse(bp);
-       free(dev, bn);
+
+       /*
+        * Recursively free last partial block.
+        */
+       if (level > SINGLE && lastbn >= 0) {
+               last = lastbn % factor;
+               nb = bap[i];
+               if (nb != 0)
+                       blocksreleased += indirtrunc(ip, nb, last, level - 1);
+       }
+       brelse(bp);
+       return (blocksreleased);

 }
 
 /*
 }
 
 /*

- * Make a new file.
+ * Remove any inodes in the inode cache belonging to dev.
+ *
+ * There should not be any active ones, return error if any are found
+ * (nb: this is a user error, not a system err).

  */
  */

-struct inode *
-maknode(mode)
+#ifdef QUOTA
+iflush(dev, iq)
+       dev_t dev;
+       struct inode *iq;
+#else
+iflush(dev)
+       dev_t dev;
+#endif

 {
        register struct inode *ip;
 
 {
        register struct inode *ip;
 

-       ip = ialloc(u.u_pdir->i_dev);
-       if(ip == NULL) {
-               iput(u.u_pdir);
-               return(NULL);
+       for (ip = inode; ip < inodeNINODE; ip++) {
+#ifdef QUOTA
+               if (ip != iq && ip->i_dev == dev)
+#else
+               if (ip->i_dev == dev)
+#endif
+                       if (ip->i_count)
+                               return (EBUSY);
+                       else {
+                               remque(ip);
+                               ip->i_forw = ip;
+                               ip->i_back = ip;
+                               /*
+                                * as i_count == 0, the inode was on the free
+                                * list already, just leave it there, it will
+                                * fall off the bottom eventually. We could
+                                * perhaps move it to the head of the free
+                                * list, but as umounts are done so
+                                * infrequently, we would gain very little,
+                                * while making the code bigger.
+                                */
+#ifdef QUOTA
+                               dqrele(ip->i_dquot);
+                               ip->i_dquot = NODQUOT;
+#endif
+                       }

        }
        }

-       ip->i_flag |= IACC|IUPD|ICHG;
-       if((mode&IFMT) == 0)
-               mode |= IFREG;
-       ip->i_mode = mode & ~u.u_cmask;
-       ip->i_nlink = 1;
-       ip->i_uid = u.u_uid;
-       ip->i_gid = u.u_gid;
+       return (0);
+}

 
 

-       /*
-        * Make sure inode goes to disk before directory entry.
-        */
-       iupdat(ip, &time, &time, 1);
+/*
+ * Lock an inode. If its already locked, set the WANT bit and sleep.
+ */
+ilock(ip)
+       register struct inode *ip;
+{

 
 

-       wdir(ip);
-       return(ip);
+       ILOCK(ip);

 }
 
 /*
 }
 
 /*

- * Write a directory entry with
- * parameters left as side effects
- * to a call to namei.
+ * Unlock an inode.  If WANT bit is on, wakeup.

  */
  */

-wdir(ip)
-struct inode *ip;
+iunlock(ip)
+       register struct inode *ip;

 {
 
 {
 

-       u.u_dent.d_ino = ip->i_number;
-       bcopy((caddr_t)u.u_dbuf, (caddr_t)u.u_dent.d_name, DIRSIZ);
-       u.u_count = sizeof(struct direct);
-       u.u_segflg = 1;
-       u.u_base = (caddr_t)&u.u_dent;
-       writei(u.u_pdir);
-       iput(u.u_pdir);
+       IUNLOCK(ip);

 }
 }